Fto in the hippocampus mediates depression-like behaviors

Dynamic and reversible RNA methylation has emerged as a new layer of epigenetic regulation of behaviors such as learning and memory; however, whether RNA methylation plays a critical role in the pathophysiology of depression is unclear. Here, we report that expression of the fat mass and obesity associated gene (FTO), a primary RNA demethylase, is downregulated in the hippocampi of both major depressive disorder (MDD) patients and mouse models of depression. Suppressing Fto expression in the hippocampus induces depression-like behaviors in mice, while elevating its expression leads to antidepressant effects. Epitranscriptomic proling of N6-methyladenosine (m 6 A) RNA methylation in the hippocampi of Fto knockdown (KD), Fto knockout (cKO), and Fto-overexpressing (OE) mice identied adrenoceptor beta 2 (Adrb2) mRNA as a target of Fto. Adrb2 stimulation reverses the depression-like behaviors and spine loss induced by hippocampal Fto deciency, possibly via the modulation of hippocampal Sirt1 expression by c-Myc. These ndings reveal that Fto in the hippocampus mediates depression-like behaviors and highlight the importance of reversible RNA methylation in driving depression. analysis to obtain the corrected P-values and fold change values. The microarray data were validated by qPCR, and the results for this microarray data have not been published. The expression of genes encoding RNA methylation-modifying enzymes was derived from the microarray data (in house data).


Background
Major depressive disorder (MDD) is currently the leading cause of disability worldwide and is characterized by low mood, diminished interests, feelings of despair or guilt, impaired cognition and other symptoms 1 . Epigenetic regulation of genes induced by environment interactions strongly contributes to the pathophysiology of depression 2 . Considerable progress in the understanding of DNA methylation and histone modi cations has been made 3 ; however, the epitranscriptomic in uences on depression remain largely elusive.
N6-methyladenosine (m 6 A) methylation, as the most abundant internal modi cation of eukaryotic mRNA, occurs extensively in the brain 4,5 and is dynamic and reversible in the mammal 6 . FTO is a wellcharacterized RNA demethylase that is expressed at the highest levels in multiple brain regions 7 . Previous studies have demonstrated that Fto in the brain regulates postnatal growth of mice 8 , the activity of the dopaminergic midbrain circuitry 9 , memory processes in the prefrontal cortex 10 and the hippocampus 11-14 , brain development 12 , adult neurogenesis 12 , axonal regeneration 15 , and localized mRNA translation 16,17 . Genetic studies have revealed that FTO is associated with depressive symptoms [18][19][20] and other studies have also showed FTO involve in the pathogenesis of MDD 11,21 . These studies showed that FTO may play a vital role in MDD, however, the speci c role of FTO in mediating depression-like behaviors and the underlying mechanisms have yet to be clari ed.

FTO expression is downregulated in the hippocampi of MDD patients and mouse models of depression
We performed microarray analysis of the peripheral blood of 36 MDD patients and 20 healthy controls, and found that both RNA methyltransferases (METTL3, METTL14, and WTAP) and demethylases (FTO and ALKBH5) were signi cantly downregulated in MDD (Fig. 1a). We further measured the expression of m 6 A-modifying enzymes in three mouse models of depression, including the chronic unpredictable mild stress (CUMS), chronic restraint stress (CRS), and social defeat stress (SDS) models, which are widely employed in preclinical studies of depression. In CUMS model mice, the mRNA expression of Fto and Alkbh5 in the peripheral blood was signi cantly downregulated compared to that in control mice, whereas the expression of RNA methyltransferases remained unchanged (Fig. 1b). Behavioral tests showed that all three models of depression were successfully established (Fig. 1c, d, g, h, k, and l). We examined the mRNA expression of m 6 A-modifying enzymes in multiple brain regions involved in regulating emotion, including the prefrontal cortex (PFC), nucleus accumbens (NAC), amygdala (AMY), and hippocampus (HIP). We found that Fto was consistently downregulated in the hippocampi of all three animal models compared to the hippocampi of control mice (Fig. 1e, i, m), whereas there was either no change or inconsistent changes in the expression of the other examined enzymes ( Supplementary Fig. 1a-c). Western blot analysis further con rmed that Fto was signi cantly downregulated in the hippocampi of all three mouse models compared to the hippocampi of control mice (Fig. 1f, j, n). Postmortem study revealed that mRNA expression of FTO was speci cally and signi cantly decreased in the hippocampi of MDD patients compared to healthy controls (Fig. 1o, Supplementary Table 1).

Fto mediates depression-like behaviors in the hippocampus
To test how hippocampal Fto contributes to depression-like behaviors, we employed recombinant adenoassociated virus (rAAV) as a transgenic tool to speci cally knock down and overexpress mouse Fto in the hippocampi of wild-type C57BL/6 mice and to speci cally knock out Fto in the hippocampi of Fto ox/ ox mice. Four weeks after bilateral microinjection of rAAV into the hippocampus (Fig. 2a), Fto protein expression was manipulated as expected in shRNA-expressing (knockdown; KD), Ftooverexpressing (OE), and Cre-expressing (cKO) mice (Fig. 2b). As Fto catalyzes RNA demethylation, the m 6 A dot blot assay showed that altering the expression level of Fto increased m 6 A levels in hippocampal total RNA from KD and cKO mice and decreased m 6 A levels in hippocampal total RNA from OE mice (Fig.   2c). Next, we analyzed the behavioral consequences of changes in Fto expression. KD mice demonstrated a longer immobility duration in the tail suspension test (TST) and a lower sucrose preference in the sucrose preference test (SPT) than control mice (Fig. 2d, e). cKO mice showed longer immobility durations in the TST and the forced swimming test (FST) as well as a longer latency to feed in the novelty suppressed feeding test (NSFT) than EGFP-expressing mice (Fig. 2f, g, h). These results reveal that suppressing Fto expression in the hippocampus is su cient to induce depression-like behaviors in mice. Subsequently, we asked whether overexpression of Fto leads to antidepressant effects. CUMS induced depression-like behaviors in EGFP-expressing mice; however, stress-treated OE mice demonstrated a shorter immobility duration in the TST and a higher sucrose preference in the SPT than CUMS-treated EGFP-expressing mice (Fig. 2i-k). These results reveal that overexpression of Fto in the hippocampus has antidepressant effects.
Recent studies have reported various effects of systemic Fto knockout on anxiety-like behaviors in mice 13,21 . Owing to the high comorbidity of depression and anxiety, we also analyzed the anxiety-like behaviors of KD and cKO mice. There was no signi cant difference in the time spent in the center zone in the open eld test (OFT) between KD and control mice ( Supplementary Fig. 2a). Additionally, in the elevated zero maze test (EZM), there was no signi cant difference in the time spent in the open arms between KD and control mice ( Supplementary Fig. 2b). Similar results were observed for cKO mice ( Supplementary Fig. 2c, d). These results are consistent with a previous study that used a speci c Cre line to knock out Fto in the CA1 and CA3 regions of the hippocampus 11 . Our study demonstrates that Fto in the hippocampus speci cally mediates depression-like behaviors without affecting anxiety-like behaviors.
Speci c deletion of Fto in the nervous system results in postnatal growth retardation 8 ; therefore, we asked whether the depression-like phenotype observed in cKO mice is due to abnormal growth. However, there was no signi cant difference in body weight between control and cKO mice, indicating that downregulation of Fto in the hippocampus does not affect growth ( Supplementary Fig. 3a). Moreover, after 24 hours of food deprivation, there were no changes in body weight between the two groups of mice ( Supplementary Fig. 3b, c).

Modi cation of the hippocampal m 6 A epitranscriptome by Fto
How does hippocampal Fto mediate depression-like behaviors? We aimed to answer this question by performing m 6 A epitranscriptomic analysis of hippocampal mRNAs by MeRIP-Seq. After quality control, 47.01 to 76.53 million reads were generated from each m 6 A-seq library. An average of 88.59% reads were uniquely mapped to the mouse genome and retained for further analysis (Supplementary Table 2).
Approximately 25% -34% of the expressed genes were modi ed by m 6 A, with approximately 2.65 peaks per gene. The pattern of m 6 A modi cation distribution on genes was similar among groups ( Supplementary Fig. 4a). Consistent with previous studies 5, 22 , the majority of m 6 A peaks were preferentially located on the 3'UTR and around the stop codon, and the enriched motifs in all samples contained 'GGAC', one of the most frequent m 6 A consensus motifs (Fig. 3a).
Next, we analyzed the differentially methylated peaks (DMPs) among different groups (Fig. 3b). Most DMPs were distributed in exons and the 3'UTR ( Supplementary Fig. 4b). KEGG pathway analysis showed that multiple gene clusters were enriched in Wnt, calcium, Notch, and axon guidance signaling pathways ( Fig. 3c). Speci cally, hypermethylated genes in cKO mice and hypomethylated genes in OE mice were both enriched in the neuroactive ligand-receptor interaction pathway. GO analysis demonstrated that genes with hypomethylated peaks in OE mice were mostly enriched in pathways regulating synapse organization and signal release, neurotransmitter secretion and transport, and neural precursor cell proliferation (Fig. 3d).
Adrb2 is functionally important target gene of FTO in the hippocampus As Fto is a well-characterized RNA demethylase, we analyzed both hypermethylated mRNAs in KD and cKO mice and hypomethylated mRNAs in OE mice (Fig. 4a). After mapping each peak to the corresponding gene, we found that 382 and 661 genes were hypermethylated in KD and cKO mice, respectively, while there were 613 hypomethylated genes in OE mice (Fig. 4b). Finally, we identi ed 34 genes that were both hypermethylated in KD and cKO mice and hypomethylated in OE mice ( Fig. 4c; Supplementary Table 3). Gene-speci c m 6 A qPCR analysis of Zfp217 and Ahnak con rmed the MeRIP-Seq results ( Supplementary Fig. 4c). At the mRNA level, Zfp217 and Ahnak were downregulated in cKO mice and upregulated in OE mice compared to control mice ( Supplementary Fig. 4d). Among these 34 genes, Adrb2 encodes the β2-adrenergic receptor, which has been reported to regulate synaptic plasticity, adult neurogenesis, and neuroin ammation in the hippocampus and to exert antidepressant effects [23][24][25][26][27] . Therefore, we wondered whether Adrb2 contributes to Fto-mediated depression-like behaviors in the hippocampus. First, we con rmed that suppressing Fto expression hyper-methylates Adrb2 mRNA in KD and cKO mice and that Fto overexpression leads to hypomethylation of Adrb2 mRNA in OE mice ( Fig. 4d, g, j). At both the mRNA and protein levels, Adrb2 was downregulated in KD and cKO mice but upregulated in OE mice compared to control mice ( Fig. 4e, h, k, f, i, and l). Postmortem study revealed that mRNA expression of ADRB2 was signi cantly decreased in the hippocampi of MDD patients compared to healthy controls (Fig. 4m). We further investigated whether changes in m 6 A methylation affect Adrb2 mRNA levels in Neuro-2a cells. In the presence of actinomycin D, an inhibitor of mRNA transcription, Fto knockdown accelerated the degradation of Adrb2 mRNA, whereas Fto overexpression signi cantly delayed this process (Fig. 4n, o). When Ythdf2, an m 6 A reader protein that regulates RNA stability 28 , was knocked down, the level of Adrb2 mRNA signi cantly increased (Fig. 4p). These results indicate that m 6 A modi cation affects the level of Adrb2 mRNA, at least partly by regulating its stability.

Adrb2 stimulation reverses the depression-like behaviors and spine loss induced by hippocampal Fto de ciency
We investigated whether Adrb2 contributes to the phenotype induced by hippocampal Fto de ciency. We tested this role of Adrb2 in cKO mice by stimulating Adrb2 with formoterol (FOR), a speci c Adrb2 agonist that can cross the blood-brain barrier (Fig. 5a). FOR administration rescued the depression-like behaviors of cKO mice in both the FST and TST in two cohorts of mice, while pretreatment with ICI 118,551 (ICI), a speci c Adrb2 antagonist, blocked the effects of FOR (Fig. 5b, c). Depression has been demonstrated to be associated with synaptic dysfunction in multiple brain regions, including the hippocampus 29 . Treatment of cKO mice with FOR for 7 consecutive days signi cantly increased the number of spines to a level comparable to that in EGFP-expressing mice, whereas pretreatment with ICI blocked the effects of FOR (Fig. 5d).
Adrb2 mediates depression-like behaviors possibly via the modulation of hippocampal Sirt1 expression by c-Myc Sirt1, an NAD + -dependent deacetylase, is essential for synaptic plasticity in the hippocampus 30,31 , and it has been reported that hippocampal Sirt1 signaling mediates depression-like behaviors 32 . In our study, the Sirt1 protein was downregulated in the hippocampi of cKO mice but upregulated in OE mice (Fig. 5e).
However, in all three m 6 A-seq libraries, we detected no change in the m 6 A level on Sirt1 mRNA, suggesting that Fto may regulate the expression of Sirt1 through other pathways. Recent studies have reported that Adrb2 activation upregulates Sirt1 expression in cervical cancer cells through promoting the expression of c-Myc 33 . In vitro, c-Myc can bind to the Sirt1 promoter and induce Sirt1 expression 34 . Thus, we wondered whether Adrb2 exerts a similar effect in the hippocampus. Blocking Adrb2 signi cantly decreased Sirt1 protein levels in the hippocampi of wild-type mice after i.p. administration of ICI (Fig. 5f). FOR treatment obviously increased Sirt1 expression in cKO mice compared to saline-treated cKO mice, while pretreatment with ICI blocked the effects of FOR (Fig. 5g). Similar results as those observed for Sirt1 were found for c-Myc in OE or cKO mice (Fig. 5h). There was also no change in the m 6 A level on c-Myc mRNA in the KD, cKO, and OE m 6 A-seq libraries. FOR treatment also obviously increased c-Myc expression in cKO mice compared to saline-treated cKO mice, while pretreatment with ICI blocked the effects of FOR (Fig. 5g). Stimulation of Adrb2 by FOR signi cantly increased c-Myc protein expression in wild-type mice compared to saline-treated mice, and this effect was blocked by pretreatment with ICI ( Fig.  5i). Finally, overexpression of Sirt1 in the hippocampi of cKO mice effectively reversed the depression-like behaviors induced by Fto de ciency ( Fig. 5j-m). These results suggest that Adrb2 acts downstream of Fto in the hippocampus to mediate depression-like behaviors and that Adrb2 may target Sirt1 through modulating the expression of c-Myc.

Discussion
Our data reveal that the expression of hippocampal FTO is decreased in both MDD patients and three mouse models of depression. Speci c knockdown or knockout of Fto in the hippocampus induces depression-like behaviors, whereas overexpressing Fto has antidepressant effects. Epitranscriptomic analysis identi ed Adrb2 as one of the targets modi ed by Fto. Loss of Fto in the hippocampus decreases the expression of Adrb2 by elevating m 6 A level in Adrb2 mRNA. Conversely, overexpression of Fto increases the expression of Adrb2 by reducing m 6 A level in Adrb2 mRNA. Adrb2 stimulation can rescue depressive-like behaviors and spine loss caused by hippocampal Fto de ciency and this effect can be reversed by Adrb2 inhibition, which may act via c-Myc to modulate hippocampal Sirt1 expression ( Fig. 6).
Genetic studies show that the FTO rs9939609 variant exerts a protective effect against depression 19 , and further analysis reveals that this positive association is driven by an atypical MDD subtype 18 . However, how Fto functions in the pathophysiology of depression and the roles it plays remain largely unknown.
Inconsistent results regarding anxiety phenotypes found in different Fto mutant mice 11,13,21 highlight the regional speci city of the functions of Fto in the brain, such as modulation of dopaminergic signaling in the midbrain and regulation of fear memory in the prefrontal cortex. We found that Fto in the hippocampus is critical in mediating depression-like behaviors.
Adrb2 is an important target gene of FTO through MeRIP-seq assay and functional studies. Although FTO may target many mRNAs, our study focused on Adrb2 mRNA because Adrb2 can regulate synaptic plasticity and contribute to many mental disorders [24][25][26]35 . Previous studies have reported that stimulation of central beta 2 -adrenergic receptors can produce antidepressant-like effects on behaviors of animals 36 . While a large of number of studies indicates that activation of central Adrb2 can produce antidepressant-like effects on animals, the speci c mechanisms remain unclear. In our study, we found that depleted expression of Fto increases m6A levels at 3'UTRs, which in turn leads to the upregulation of Adrb2 at the RNA and the protein level. Conversely, m6A level was decreased in the hippocampal tissues of Fto overexpression and the expression of Adrb2 was increased. Wang et.al reported that m6Acontaining mRNA intends to be less stable because such mRNA can be relocalized to the decay sites by YTHDF2 28 . Our results showed that Fto knockdown can accelerate the degradation of Adrb2 mRNA, and overexpression of FTO can alleviate the degradation of Adrb2 mRNA. Taken together, the above results revealed that FTO regulates the expression of Adrb2 relying on demethylase activity to a certain extent.
Two loci, one of which is near the SIRT1 gene, were identi ed as the rst robust links between genetics and depression at the whole-genome level 37 . Recent studies reported the expression of Sirt1 is signi cantly less in MDD patients compared to healthy subjects in the peripheral blood 38 , which is consistent with our microarray analysis (in house data). Animal studies also show that Sirt1 regulates depression-like behaviors in several brain regions, including the hippocampus 32,39−41 . We observed that the methylation level of Sirt1 mRNA is not modi ed by Fto; however, the protein expression of Sirt1 is altered by Fto. Chen et al found that the expression of Sirt1 can be upregulated through activating Adrb2 through elevating c-Myc expression in cervical cancer cells 33 . c-Myc is reported to be a transcriptional factor of Sirt1 and can regulate the expression of Sirt1 in vitro 42 . Consistent with previous results, we found that injection of FOR can increase the expression of c-Myc protein in the hippocampus and injection of ICI can blocked this effect. In addition, the expression of c-Myc protein was decreased or increased in the hippocampus of FTO knockout or overexpression. For the rst time, we report that Adrb2 regulates Sirt1 expression in vivo in the hippocampi of wild-type and Fto-de cient mice. Both Adrb2 activation and elevation of hippocampal Sirt1 levels reverse the depression-like behaviors induced by Fto de ciency, suggesting that Fto may target Sirt1 through modifying the methylation of Adrb2 mRNA.
This study indicates that Fto in the hippocampus mediates depression-like behaviors and might be a new potential target for depression therapy. Recently developed speci c FTO inhibitors show promise for treating cancer 43 and obesity 44 . Our ndings also suggest that based on evidence from several weightloss drugs 45 , attention should be paid to potential adverse psychiatric events during systematic administration of FTO inhibitors.

Human subjects
Thirty-six unrelated MDD patients (20 females and 16 males) aged 39.9 ± 12.5 (SD) years and twenty unrelated healthy subjects (10 females and 10 males) aged 36.6 ± 11.1 (SD) years were recruited from the DBPR (manuscript NCOMMS-21-04052). All subjects provided written informed consent prior to participating in this study, which was approved by the Ethics Committee of the DBPR (manuscript NCOMMS-21-04052). None of the MDD patients or healthy controls had taken any psychotropic medications within 4 weeks. All patients were clinically diagnosed by at least two psychiatrists according to the Diagnostic and Statistical Manual of Mental Disorders Fourth Edition (DSM-IV) criteria and the 17item Hamilton Rating Scale for Depression (HAMD-17). Hippocampal tissues of the human brain were provided by the Human Brain Bank, DBPR (manuscript NCOMMS-21-04052). The postmortem intervals between the death of donors and brain collection were less than 40 h, and detailed information on the donors is provided in Supplementary Table 1.

Microarray
Total blood RNA was isolated with a PAXgene Blood RNA Kit (Qiagen, 762174). The optical density values at 260/280 were approximately 1.9-2.0, and the quality of the RNA was also con rmed with an Agilent 2100 (Agilent Technologies, Santa Clara, CA). RNA samples underwent reverse transcription to synthesize double-stranded complementary DNA (cDNA), and the cDNA was then transcribed into cRNA in vitro. The cRNA was then reverse-transcribed into cDNA. The cDNA was uorescently labeled and hybridized using a 4 × 180K lncRNA+mRNA Human Gene Expression Microarray V3.0 (CapitalBio Technology, Beijing, China). Agilent Feature Extraction (v10.7) software (Santa Clara, CA, USA) was used to analyze the hybridization map and extract the data. The chip uorescence scanning images were saved as DAT les for analysis by the Agilent G2565CA Microarray Scanner. The data normalization and difference analysis were performed with Agilent GeneSpring software. Two groups of sample data underwent a t-test analysis to obtain the corrected P-values and fold change values. The microarray data were validated by qPCR, and the results for this microarray data have not been published. The expression of genes encoding RNA methylation-modifying enzymes was derived from the microarray data (in house data). The Neuro-2a cell line, which was provided by the Cell Resource Center (IBMS, CAMS/PUMC), was cultured in DMEM supplemented with 10% heat-inactivated fetal bovine serum (04-001-1A, Biological Industries), 100 U/ml penicillin and 100 μg/ml streptomycin at 37 °C in 5% CO 2  Chronic restraint stress (CRS). CRS exposure was performed as previously described 47 . Mice were placed in polypropylene conical tubes (50 ml) with pores at the bottom enabling air ow. Animals had no access to food and water during restraint stress. Four hours later, the mice were returned to their home cages. The same stressor was administered once at a random time each day, which was continued for 21 days.
Social defeat stress (SDS). Social defeat stress exposure was performed according to the published protocol 48 . Brie y, C57BL/6J mice were placed into a cage in the presence of CD-1 aggressor mice for 5-10 min per day for 10 consecutive days. Then, the C57BL/6J and CD-1 mice were separated by a perforated Plexiglas plate for 24 h, which blocked all physical contact. Each C57BL/6J mouse was exposed to different CD-1 mice every day.

Behavioral test
Animals were transported and acclimated to the testing room 1 h before each behavioral test.
Forced swimming test (FST). Mice were individually placed in a beaker (height: 19 cm; diameter: 14 cm) containing 14 cm of water (23 ± 2 °C). The total test duration was 6 min. The process was videotaped, and the immobility time in the last 4 min was scored by an experienced observer who did not know the experimental treatment. Floating or only slight movement to maintain balance was considered immobility.
Tail suspension test (TST). The TST was performed with tail suspension hardware (Med Associates, USA). The mouse was suspended by its tail with a short adhesive tape connected to a load cell that transmits a signal corresponding to activity. The total test time was 6 min. After setting a lower threshold, the duration of immobility was recorded and analyzed by tail suspension software (SOF-821, Med Associates).
Sucrose consumption test (SCT). Mice were acclimatized to 1% sucrose solution for 48 h before the test. Then, each mouse was individually housed and deprived of water for 12 h. SCT was performed for 1 h, in which the amount of the 1% sucrose solution consumed was recorded.
Sucrose preference test (SPT). Mice were acclimatized to two-bottle drinking (one bottle contained water, and the other contained a 1.5% sucrose solution) for 48 h, during which the position of each bottle was switched at 24 h. Then, the mice were deprived of drinking water for 24 h. SPT was performed by providing the mice once again with two bottles for drinking for 2 h, during which the position of each bottle was switched after 1 h. The ratio of the sucrose solution consumed to the total uid intake in the SPT was determined as the sucrose preference.
Social interaction test (SIT). The SIT was performed according to a previous report. It consists of two 2.5 min phases: the rst is performed without a CD-1 mouse, and the second is performed with a CD-1 mouse. The two phases were separated by a pause of 30 sec. In both phases, the C57BL/6J mouse was placed into the rear corner of the open eld opposite the wire-mesh compartment. The time spent in the interaction zone (14 cm × 24 cm) was recorded and analyzed by the EthoVision video tracking system (Noldus, the Netherlands). The social interaction ratio was calculated as the time spent in the interaction zone in the presence of the CD-1 mouse to the time spent in the absence of a CD-1 mouse. Brains of susceptible and control mice were harvested for gene expression analysis.
Novel suppressed feeding test (NSFT). We performed the NSFT by referring to a previous study. Mice were deprived of food for 24 h. During the NSFT, each mouse was placed into the corner of a square chamber (50 cm × 50 cm× 20 cm) opposite a food pellet that was placed in the center of the chamber.
The latency to start eating the pellet was recorded. If the mouse did not eat the pellet within 6 min, the latency was recorded as 6 min.
Open eld test (OFT). Each mouse was individually placed in one corner of a square Plexiglas box (50 cm × 50 cm × 40 cm) in a brightly lit room, allowing for free exploration of the open eld arena for 10 min. The time spent in the center area of the box (12.5 cm × 12.5 cm) and the distance traveled in the box during the testing period were recorded and analyzed by the EthoVision video tracking system (Noldus, the Netherlands).
Elevated zero maze test (EZM). The EZM was performed in a circular elevated maze with two open and two enclosed arms (height: 43.5 cm; diameter: 45 cm; track width: 6.5 cm). Each mouse was individually placed in the cross area of the maze facing the closed arms and allowed to freely explore the maze for 5 min. The time spent in the open arms and the distance traveled in the maze were recorded and analyzed by the EthoVision video tracking system (Noldus, the Netherlands).
Four weeks later, further experiments were performed. m 6 A Dot blot Total RNA in 3 sample volumes of RNA incubation solution was denatured at 65 ℃ for 5 min and immediately chilled on ice. RNA samples were spotted in triplicate onto an Amersham Hybond-N+ membrane (RPN119B, GE Healthcare) with a Bio-Dot Apparatus (#170-6545, Bio-Rad). Next, they were crossed-linked to the membrane by UV light and then blocked with blocking buffer (PBST buffer containing 5% nonfat milk). The membrane was incubated overnight at 4 °C with anti-m 6 A antibody (1:5000, 202003, Synaptic Systems) followed by incubation with HRP-conjugated goat anti-rabbit IgG (1:5000, sc-2030, Santa Cruz Biotechnology) for one hour at room temperature. Finally, the signal was detected by a chemiluminescent reaction using an ECL reagent.

MeRIP-seq
A re ned MeRIP-Seq (N6-methyladenosine (m 6 A) RNA immunoprecipitation sequencing) method was performed based on a previously described protocol 49,50 . The hippocampus was dissected from 3 mice, and the tissues were pooled together to obtain a su cient quantity of mRNA for immunoprecipitation followed by RNA sequencing. Brie y, poly-A-puri ed RNA was extracted and randomly fragmented into approximately 100-200 nucleotide-long fragments. After enriching for methylated RNA fragments with a speci c anti-m 6 A antibody (1:100, ABE572, Millipore), the m6A-seq library was constructed using the SMARTer Stranded Total RNA-Seq Kit-v2 (Pico Input Mammalian, 634413, Takara/Clontech, Japan) and sequenced on Illumina Novaseq 6000.
Bioinformatics analysis of high-throughput sequencing data Alignment. Sequencing data were analyzed according to the optimized pipeline for re ned MeRIP-Seq described in a previous study 50 . The quality control of raw data was performed by FASTQC (version 0.11.5). The clean reads from the m 6 A IP (m 6 A-seq) were aligned to the mouse genome (mm10) using the STAR aligner (version 2.7) with the default parameters after removing reads containing adapters and lowquality reads with cutadapt (version 1.18). Only uniquely mapped and unduplicated reads remained for the subsequent analysis. m 6 A peak calling and motif analysis. MeTPeak (version 1.1) with the default settings was used for the transcriptome-wide detection of m6A sites in each sample based on the alignment les generated by STAR aligner. The R package Guitar was used to illustrate the distribution of m6A peaks in mRNA 51 . The top 3000 peaks were selected for motif analysis with DREME (version 5.1.1), which used 100-nt-long peak summit-centered sense sequences as input. Peaks were annotated to overlapping genomic features and visualized with ChIPseeker (version 1.22.1) using annotations from TxDb.Mmusculus.UCSC.mm10.knownGene. Peaks were assigned to seven regions, including the promoter, 5' UTR, 3' UTR, exon, intron, and downstream and distal intergenic regions.
Detection of differential methylation peaks (DMPs). DMPs were identi ed according to the pipeline described in the four groups included in our study, including wild-type (WT), KD, cKO and OE mice. The following steps were taken to identify DMPs between three groups based on the expression levels in Ftomodi ed and control samples (WT mice). First, the overlapping m 6 A peaks from two samples were de ned as common peaks. The common peaks between samples from the same group were merged and combined with the common peaks from another group to form the m 6 A union peaks reference list. Then, the read counts for the union peaks for the IP and input samples were calculated separately and corrected based on the total number of ERCC spike-in reads. The enrichment score of each sample for each peak was obtained by dividing the read counts for IP by the input. High con dence DMPs meeting the criteria of an average enrichment score higher than 6 in the groups and a log2 fold change > 1 or < −1 were selected for further analysis. The DMPs were annotated according to the genomic features with ChIPseeker (version 1.22.1) 52 . Functional enrichment analysis of the gene ontology (GO) and pathways was performed using ChIPseeker (version 1.22.1) to identify the predominant biological functions of the annotated genes nearest to the DMPs. Fisher's exact test was applied to identify the signi cant GO categories and pathways, and FDR correction was performed with a threshold of 0.05.

Gene-speci c m 6 A qPCR
The Magna MeRIP m 6 A kit (17-10499, Millipore) was used to perform m 6 A RNA immunoprecipitation according to the manufacturer's instructions. A QuantiTect SYBR Green RT-PCR Kit (204243, Qiagen) was used to perform reverse transcription and real-time qPCR. The relative expression of the m 6 A-modi ed target gene was determined as the Cq value of the m 6 A IP portion divided by the Cq value of the input portion.

RNA stability
Neuro-2a cells were transfected with Fto-shRNA and Fto overexpression and control vectors. Forty-eight hours later, the cells were treated with 5 μg/ml actinomycin D (A9415, Sigma) and harvested at 0, 3 and 6 h. Total RNA was extracted using TRIzol, and 1 μg of total RNA was used for reverse transcription and qPCR.

Golgi staining
Mice were anesthetized with chloral hydrate (10%, wt/vol, i.p.) and sequentially perfused transcardially with saline. The whole brain was freshly harvested and was immediately immersed in one well of a 12well plate containing 5 ml of solution A (superGolgi Kit, Cat. # 003010, Bioenno Tech, LLC). According to the manufacturer's instructions, the solution was renewed two days later, and the impregnation continued for another 9 days. Next, the brain was transferred to postimpregnation buffer for 2 days with renewal of the solution after one day of immersion. Sections that were 100 μm in thickness were cut using a vibratome (VT1200S, Leica). The sections were mounted on adhesive microscope slides (SUPERFROST® PLUS, Cat. # 4951 PLUS-100E, Thermo Scienti c). The staining time in solutions C and D was 20 min.
Then, the sections were dehydrated in 100% ethanol for 7 min 4 times each, which was followed by vitri cation with xylene for 8 min 10 times each. Finally, the sections were coverslipped with neutral resins. For spine density measurement, we imaged one terminal dendrite from the CA1 region of the hippocampus using a 100× objective on a microscope (DM6 B, Leica). A Z-stack of images with an interval of 0.22 μm was acquired for each dendritic segment, which was followed by an extended depth of eld algorithm calculation to obtain in focus images using LAS X software (Leica Application Suite X, Leica). The diameters of the selected segments were comparable. A total of 120 segments from 20 mouse brains (6 segments per animal) were used to determine the spine number. The dendritic spines, including thin, stubby, and mushroom types, within a 15-μm length segment were counted manually by an investigator blinded to the experimental treatments.

Statistical analysis
Statistical signi cance was analyzed with Prism software (GraphPad Prism 8). Two-tailed student's t test was used for comparisons between the two groups, and one-way or two-way ANOVA was performed for more than two groups. P < 0.05 was considered statistically signi cant. All the data were exported into